Assessing the Validity and Reliability of a French Version of the Child Feeding Questionnaire among Black Mothers Living in Ottawa (Canada)

The Child Feeding Questionnaire (CFQ) is a well-established tool used to assess parental beliefs, attitudes, and child feeding practices, with a focus on childhood obesity proneness. To date, there is no French version of the CFQ, nor any Canadian studies assessing its construct validity. The aim of this study was to assess the construct validity and the reliability of a French version of the CFQ among Black mothers (n = 136) of school-aged children living in Ottawa (Canada). The final best fitting model included 7 factors, 20 items, and 1 error covariance. This model was retained as the final model as it (1) excluded two items with very low factor loadings; (2) had the lowest χ2, AIC, BIC, RMSEA, and SRMR values; and (3) had CFI and TLI values ≥ 0.95. Internal consistency ranged from poor to good; the restriction subscale had the lowest internal consistency, followed by the perceived responsibility, pressure to eat, perceive child weight, concern about child weight, and monitoring scales, respectively. Our results showed that a seven-factor model with minor modifications was best fitted to the current data. Future studies are needed to test the validity and reliability of the CFQ in other population groups and among fathers.


Introduction
Childhood obesity is a growing public health problem involving both immediate and long-term consequences on physical (e.g., higher risk of developing type 2 diabetes, cardiovascular disease) and mental (e.g., lower self-esteem, higher risk of developing depression) health [1,2]. It is well-established within the scientific literature that, due to differences in genetic backgrounds, health-related behaviors, and socioeconomic status, the prevalence of excess weight varies across racial and ethnic groups [3,4]. In Canada, Black individuals are among the highest-risk population groups, with overweight and obesity rates nearing 50% and 29% among adults and children (2-17 years old), respectively [3,5,6].
Socioecological factors such as family have been shown to influence childhood obesity risk [7,8]. According to Anderson et al., "family is the first and most fundamental socioenvironmental context in which eating patterns are established" [9]. Through shaping the family's food environment, parents influence their child's eating behaviors and perhaps their lifelong weight trajectories [10,11]. Studies assessing parental child feeding practices have reported that controlling child feeding practices such as restriction and pressure to eat could interfere with the child's ability to recognize internal hunger and satiety cues, which could eventually impact the child's weight status [12][13][14][15]. In this regard, several crosssectional and longitudinal studies have reported that parents who use restrictive feeding practices (i.e., restricting their children's access to foods, particularly "unhealthy" foods)

Participants
Mothers in this analytic sample were part of a cross-sectional study aiming to assess the social determinants of nutritional heath of 6-to-12-year-old children of African and Caribbean descent. The study, including recruitment strategies and findings, has been described in several previously published manuscripts [18,[41][42][43][44][45]. The study was conducted between January 2014 and April 2015 in Ottawa (Canada). To be eligible for the study, mothers had to have at least one child aged between 6 and 12 years, to be born in sub-Saharan Africa, the Caribbean, or Canada, and be able to understand and speak English or French. Due to genetics being a risk factor for overweight and obesity, mothers of adopted children were excluded. Children who had food allergies occasioning important dietary restrictions, and/or suffered from metabolic or congenital disorders that may influence growth, and/or limited the intake of certain foods, and/or suffered from any medical condition limiting or preventing their participation in normal physical activities were also excluded. When mothers had two or more children eligible to participate in the study, one of them was randomly chosen. Mothers completed the CFQ in either English or French, according to their preference. A total of 252 mothers participated in this study. Two hundred and twelve (n = 212) mothers were Black, among which one hundred and forty-one chose to complete the CFQ in French. A total of 5 mothers were excluded because they self-identified with more than one ethnicity, leaving 136 mothers for the current analyses.
The study was approved by the Office of Research Ethics and Integrity of the University of Ottawa (H-02-19-1405). Informed consent and assent were obtained from mothers and children, respectively.

Sociodemographic Characteristics
Mothers' (i.e., age, migration status, length of time since immigration, education level, and marital status) and children's (i.e., age and sex) socio-demographic characteristics were collected through a questionnaire administered to the mother by a trained research team member. Length of time since migration and age were calculated using the data collection date, along with the reported date of arrival in Canada and mothers' and children's dates of births, respectively.

Racial and Ethnic Identity
Ethnicity refers to the shared understanding of the historical and territorial origins of a population group as well as to a distinctive set of cultural similarities between people [46,47]. The latter concept is flexible and requires self-identification to one or more pre-established and/or self-perceived ethnic groups. In this study, ethnic identity was determined according to participants' responses to the question: "With which racial or cultural group do you identify? You may belong to one or more racial or cultural groups on the following list: White, South Asian, Chinese, Black, Filipino, Latin American, Arab, Southeast Asian, West Asian, Korean, Japanese, and Other".

Weight Status
Mothers' and children's height (Charder HM200P Portstad portable stadiometer) and weight (LifeSource ProFit UC-321, A&D Medical digital scale) were measured by a registered dietitian or a trained dietetic student using the World Health Organisation (WHO) guidelines [48]. Mothers' body mass index (BMI, weight (kg)/height (m) 2 ) was calculated and classified according to the WHO's BMI cut-offs [48]. The International Obesity Task Force (IOTF) references were used to calculate BMI-for-age-and-sex z-scores and define children's weight status [49].

Maternal Feeding Practices, Perceptions, and Concerns about Child's Weight
The CFQ was used to assess maternal feeding practices, perceptions, and concerns about child's weight [25]. All of the CFQ's subscales, subscales' items, and answer scales are shown in Table 2. Three items of the PPW subscale and three items of the PCW subscale were dropped prior to the study in order to shorten the survey. Items assessing mothers' perception of their weight during childhood (PPW1), adolescence (PPW2), and her twenties (PPW3) were dropped to shorten the questionnaire and due to the fact that an important proportion (94.1%) of Black mothers in this study were immigrants and thus spent part of their life in a Sub-Saharan African or Caribbean country, where rates of overweight and obesity were low. As there are important variations in body shape standards across cultures and because immigrant mothers came to Canada at different life stages, only mothers' perception of their current body weight (PPW4) was assessed. As for the PCW subscale, items assessing parents' perception of their child's weight after kindergarten (PCW4-6) were dropped. As participating children were aged 6-to-12 years old, only items PCW1-PCW3 could be answered by all mothers, while items PCW4-PCW6 could only be answered by mothers with older children.

Translation of the Child Feeding Questionnaire
The back-translation method with bilingual testing was used for translating into French the original English version of the CFQ [50]. Two bilingual (French-English) individuals whose first language was French individually completed the English-to-French translation of the CFQ. A panel composed of two bilingual co-investigators of the study compared, discussed, and modified translations to get the first complete French version of the questionnaire. Using the same approach, the French questionnaire was translated back into English by two bilingual individuals whose first language was English. The same panel compared the different English translations as well as the French and English translations, which led to a second version of the questionnaire in both languages. Both the English and the French versions of the questionnaire were tested among bilingual individuals of the target population before some final modifications were made. After the translation process, the response options "seldom", "markedly underweight", "markedly overweight", and "unconcerned" of the original English version of the CFQ were replaced by "rarely", "very underweight", "very overweight", and "not concerned", respectively, to lower the language level as many participants had low English literacy. The English and the French versions of the questionnaire are presented in Table 2.

Software and Packages
Data processing and statistical analyses were computed in R version 4.0.2 for (Mac) OS X [51]. Multivariate imputation via chained equations, item measures and reliability, and confirmatory factor analyses were computed using MICE [52], psych [53], and lavaan [54] packages.

Item Measures
All items' and subscales' mean, standard deviation, skewness, and kurtosis measures were computed as indicators of data distribution. Floor and ceiling effects were assessed and reported when an item was rated with the lowest or highest possible score, respectively, by ≥50% of participants. Spearman's ρ was computed to assess inter-item as indicators of item redundancy. Spearman's ρ values ranging between 0.15 and 0.50 for inter-item correlations were considered acceptable [55,56].

Validity
Confirmatory factor analysis (CFA) was used to evaluate the factor structure of the CFQ using maximum likelihood with robust standard error and a Santa-Bentler scaled test statistic (MLM) estimation [57]. MLM estimation was used as items evaluated on Likert-type scales can be seen as ordinal variables and may violate the assumption of multivariate normality [57]. The structures of all tested models are summarized in Table 3. Composite items present in some models were computed through averaging items scores (i.e., RST1a, RST1b, and RST1c into RST1; RST3a and RST3b into RST3; RST4a and RST4b into RST4).   The adequacy of models' fit was assessed using the χ 2 test, Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Comparative Fit Index (CFI), Tucker-Lewis Fit Index (TLI), Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Square Residual (SRMR). The χ 2 test compares the correlation matrix of the tested model to that estimated under the hypothesized (or "best fitting") model or floor effects as well as high kurtosis and skewness values, which may have contributed to the low reliability of those subscales [58]. Small and non-significant χ 2 values are indicative of small discrepancies between the observed data structure and the data structure of the hypothesized model [58]. AIC and BIC estimate the risk of under-fitting (AIC) and over-fitting (BIC) based on several parameters, including model complexity. Lower AIC and BIC values are associated with a lower risk of under-fitting and over-fitting, respectively [58][59][60]. CFI and TLI compare the tested model to a null (or "worst fitting") model, with values ranging from 0.90 to 0.94 indicating an acceptable fit and values ≥ 0.95 indicating a good fit [58,61]. RMSEA compares the tested model to a reasonably fitted one, with values ranging from 0.06 to 0.08 indicating an acceptable fit and values < 0.06 indicating a good fit [58,61]. SRMR compares the tested model to a perfectly fitted one, with values nearing 0.08 indicating a good fit [58,61].

Reliability
Cronbach α and Macdonald's ω were computed as estimates of subscales' internal consistency [62][63][64]. Cronbach α and Macdonald's ω if item deleted were also computed to identify any item that disproportionately impacted reliability estimates.

Sample Characteristics
Mothers and children were aged 39.1 (SD: 5.9, range: 28-54) and 8.8 (SD: 2.0, range: 6-12) years, respectively. A majority of mothers were foreign-born (94.1%), had a partner at the time of the study (66.2%), and obtained at least a college certificate or diploma (88.9%). Migrant mothers had been living in Canada for an average of 8.8 (SD: 8.1, range: 0.03-38.6) years. Overweight and obesity prevalence was high in both mothers (82.3%) and children (35.1%). Mothers' and children's characteristics are shown in Table 4. Table 4. Mothers' and children's characteristics.
s ω were computed as estimates of subscales' internal con-Macdonald's ω if item deleted were also computed to idenely impacted reliability estimates.

Item Measures
Items' and subscales' mean, standard deviation, distribution, and reliability estimates are shown in Table 5. There were less than 1% missing data. Skewness values were acceptable n = 128;

Item Measures
Items' and subscales' mean, standard deviation, distribution, and reliability estimates are shown in Table 5. There were less than 1% missing data. Skewness values were acceptable n = 130; ¥ n = 134.

Item Measures
Items' and subscales' mean, standard deviation, distribution, and reliability estimates are shown in Table 5. There were less than 1% missing data. Skewness values were acceptable (skew = −1.82-1.17) for the large majority of items and subscales, with only the PR3 item being negatively skewed (skew = −2.27). Several items (PR1, PR3, PCW1-3, RST1, PE2) and one subscale (PCW) had a leptokurtic distribution, as indicated by kurtosis values > 2.0. All high kurtosis values were positive, which is consistent with the fact that ceiling effects were observed for seven items; six items (PR1, PR3, RST1, RST2, RST4, PE4) were rated with the highest possible score by 50-74% of participants, and one item (PE2) was rated with the highest possible score by ≥75% participants. Floor effects were observed for CCW1 and CCW2, with these items being rated with the lowest possible score by 50-55% of participants.

Reliability
With the final model's (model 6) structure, internal consistency ranged from poor (Cronbach α = 0.53, Macdonald's ω = 0.60) to good (Cronbach α = 0.80, Macdonald's ω = 0.82) (see Table 4). The RST subscale had the lowest internal consistency (Cronbach α = 0.53, Macdonald's ω = 0. Removing any of the three items of model 6's RST subscale did not increase internal consistency. Including the composite item RST3 in the RST subscale was associated with a slightly poorer internal consistency (Cronbach α = 0.51, Macdonald's ω = 0.57). On the contrary, models without composite items on the RST scale had higher internal consistency (Cronbach α = 0.66, Macdonald's ω = 0.79). Similar to the RST scale, removing any item from the PR subscale did not improve internal consistency. For the two subscales with an acceptable internal consistency, removing an item increased Cronbach α values. Removing item PE1 from the PE subscale increased the Cronbach α value from 0.64 to 0.70, and removing item PCW3 from the PCW subscale increased the Cronbach α value from 0.68 to 0.73. The RW subscale that was only included in models 7, 8, and 9 had poor internal consistency (Cronbach α = 0.32, Macdonald's ω = 0.60).

Discussion
This study aimed to develop a French version of the CFQ and to assess its construct validity and reliability in a sample of Black mothers of 6-to-12-year-old children living in Canada. In addition to the English-to-French translation, six items (PPW1-3, PCW3-6) were removed from the original version of the CFQ prior to its administration. Based on previous validation studies, several CFQ structures were tested, among which a sevenfactor model, with minor modifications, was found to have the best fit to the current data. Most modifications made to the questionnaire's structure involved the RST subscale, whose construct limitations were highlighted in several previous studies. Except for RST and PR, all subscales had at least an acceptable level of reliability.
Similar to Schmidt et al. (2017), item analyses revealed a low number of missing data indicating that, when translated into French, the questionnaire was comprehensible overall and well-accepted by the mothers. A leptokurtic distribution and ceiling effects were noted for several items, namely, items of the PR, RST, and PE scales. While floor and ceiling effects are more likely to be observed in population-based samples [34,65], these findings raise concerns over the sensitivity of some factors of the CFQ. Since only Schmidt et al. (2017) have reported item measures, additional studies doing so are needed in order to adequately document the sensitivity of the CFQ in various population-based and clinical (e.g., treatment-seeking) samples.
In the present study, modifications made to the questionnaire during the validation process reflected previously encountered challenges. The factor causing the most problems in all validation studies is RST. While most of the items loaded as predicted and had acceptable loading factors, the composite item "food as reward" loaded poorly on the RST subscale. In the large majority of previous validation studies, these two items were either dropped [9,31,36] or used to create a new subscale entitled "food as reward" [26,32,34]. Here, we tested both approaches and retained the model in which the two RST items were dropped (Model 6, Table 3) rather than the eight-factor model (Model 9, Table 3). While both models had an acceptable construct, the retained model had better fit indices. Overall, the final model's fit indices were better than those reported in previous studies (see Table 1 for a summary of previous studies' fit indices).
Another problematic factor was PCW, which had high inter-item correlations. As previously pointed out, the repetitive nature of the questions combined with a relatively short age span may have led to strong correlations between PCW items [31,36]. To address this issue, we included one error covariance between PCW1 and PCW2 in the final model. Most previous studies, including Birch et al. (2001), also included error covariances between items of the PCW subscale in their final model to address this problem [25,31,33,34,36].
Our results showed that PCW, CCW, PE, and MN subscales had good or acceptable reliability, while PR and RST subscales had low reliability. Both the PR and the RST subscale included several items with either ceiling or floor effects as well as high kurtosis and skewness values, which may have contributed to the low reliability of those subscales [63,64]. Overall, reliability was weaker than in previous studies, which may partly be explained via differences in sample sizes. Larger sample sizes and continuous data allow for a more valid estimate of reliability [62][63][64].
This is the first study to assess the reliability and construct validity of a French translation of the CFQ. The results of this study should be considered in light of several limitations. This study was conducted in a relatively small sample size of Black mothers of 6-to-12year-old children living in Ottawa, which limits its generalizability to other racial and ethnic groups as well as to Black mothers living outside Ottawa. Another limitation of the present study is the fact that only one item from the PPW subscale was included in the questionnaire administered in the present study. Previous studies have reported high correlations between PPW items, which affect the fit indicators of the tool. The absence of these factors may have contributed to improving the fit indicators in this study. Future studies are needed to test the validity and reliability of the questionnaire among fathers, among individuals from other racial and ethnic groups, as well as among individuals living in different regions of Canada.  Informed Consent Statement: Informed consent and assent were obtained from mothers and children, respectively.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author. The data are not publicly available due to privacy restrictions.